Process mining encompasses the research area which is concerned with knowledge discovery from event logs. One common process mining task focuses on conformance checking, comparing discovered or designed process models with actual real-life behavior as captured in event logs in order to assess the fitness, precision and generalization capabilities of the process model. In many cases, such conformance checking techniques involve some kind of “replay” of the execution traces on the process model at hand. In this report, we discuss in depth the problem of replaying event traces on Petri nets for sequences containing both positive and negative events. Negative events are activities which should be prevented from taking place (contrary to opposite events which should be allowed) and are leveraged by existing conformance checking techniques to determine whether a given model is not too overfitting or underfitting. However, such negative events must be treated differently from their positive counterparts during trace replay; different replay strategies exist which impact the manner by which positive and negative events are evaluated, so that these different strategies also influence the outcome of a conformance checking evaluation. Therefore, we aim to provide an overview of the root causes which make trace replay a hard problem, together with a description of different replay strategies and their impact on process model quality evaluation.This technical report serves as background material for the following paper: vanden Broucke, S.K.L.M., De Weerdt, J.,Vanthienen, J., Baesens, B. (2013). Determining Process Model Precision and Generalization with Weighted ArtificialNegative Events. IEEE Transactions on Knowledge and Data Engineering.
展开▼
